Modeling And Prediction Of Fractional-Order Chaotic Lorenz System Using RNN And LSTM Networks

The complexities inherent underlying in the chaotic systems have made long term prediction impossible due to their sensitivity dependence on initial condition. There is need to employ machine learning to detect intricacies as they can capture patterns in complex system and also, extract fractional order behaviour from data. In this study, the comparison between the performance of Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) networks in forecasting fractional-order Lorenz chaotic time series data was investigated. The results show that training and test data for LSTM networks have lower Root Mean Squared Error (RMSE) values than the RNN values, indicating superior generalization to unseen data. By effectively modeling long-term dependencies of the chaotic system, LSTM enhances prediction accuracy and performance compared to traditional RNNs. Accordingly, these findings imply that LSTM networks are more capable of modeling fractional-order dynamics, chaotic systems, thus being more valuable in applications.